Lost and Found systems often struggle with inefficient item retrieval due to reliance on textual descriptions. This paper proposes a novel approach utilizing TensorFlow.js for image recognition within a Lost and Found systems. Our system leverages MobileNet, a pre-trained image classification model, to generate image embeddings for lost items. When a user submits an image of a found item, the system retrieves similar items from a database based on the calculated Euclidean distance between their embeddings. We present the system architecture, detailing the image processing pipeline and embedding generation process. Furthermore, we discuss the methodology employed to evaluate the system’s performance, including the calculation of distance metrics and the filtering of retrieved items based on a user-defined similarity threshold. The results are visualized to demonstrate the effectiveness of the system in identifying similar items. Finally, we compare our approach to existing text-based Lost and Found systems, highlighting the advantages of image recognition in improving retrieval accuracy. We conclude by discussing the potential for future advancements, such as incorporating user feedback mechanisms and exploring alternative image similarity metrics.